diff options
| author |  Antonio Sanchez <cantonios@google.com> | 2020-12-01 16:28:48 -0800 |
|---|---|---|
| committer |  Rasmus Munk Larsen <rmlarsen@google.com> | 2020-12-02 01:29:34 +0000 |
| commit | 2627e2f2e6125cf09fa32789755135e84552275b (patch) | |
| tree | 98aac84d6faf17d02688a2a4478c4c582932a812 /Eigen/src/Core/IO.h | |
| parent | ddd48b242cbf3aa79ad13668b66089cece6d1ea0 (diff) | |
Fix neon cmp* functions for bf16.
The current impl corrupts the comparison masks when converting
from float back to bfloat16. The resulting masks are then
no longer all zeros or all ones, which breaks when used with
`pselect` (e.g. in `pmin<PropagateNumbers>`). This was
causing `packetmath_15` to fail on arm.
Introducing a simple `F32MaskToBf16Mask` corrects this (takes
the lower 16-bits for each float mask).
Diffstat (limited to 'Eigen/src/Core/IO.h')
0 files changed, 0 insertions, 0 deletions